Trade shocks and the nationalist backlash in political attitudes: panel data evidence from Great Britain

ABSTRACT

This article leverages individual-level panel data on nationalist attitudes to contribute to the debate on the (economic) roots of popular opposition to globalization. We propose a ‘nationalist backlash’ hypothesis: Individuals living in regions suffering from stronger import competition form more nationalist attitudes as part of a broad counter-reaction to globalization. Analyzing data from the British Household Panel Study (BHPS), we document not only a decrease in support for EU membership but also a general shift towards more nationalist attitudes among respondents from regions exposed to higher imports from low-wage countries—in particular, China. We thus uncover a direct individual-level response to import shocks in the form of rising nationalist attitudes that helps to explain these shocks’ aggregate electoral consequences in terms of increased vote shares for the radical right.

KEYWORDS:

• EU support
• globalization
• import competition
• international trade
• nationalist attitudes
• panel data

1. Introduction

In the aftermath of the election of Donald Trump, the June 2016 Brexit referendum and the parallel electoral gains of right-wing populist parties in several Western democracies, scholars have vividly debated the sources behind this nationalist backlash against globalization (e.g., Hobolt, 2016; Norris & Inglehart, 2019; Rodrik, 2021; Walter, 2021). This debate has often been framed around the question of culture vs. economics: Is public opposition to globalization rooted in deep-seated cultural values—such as authoritarianism, xenophobia or nationalism—or is it best understood via economic grievances among globalization’s material losers? Yet, culture and economics might be more interconnected than implied by this simple juxtaposition (Colantone & Stanig, 2019; Gidron & Hall, 2020). Indeed, evidence begins to accumulate that cultural values are affected by economic distress experienced by individuals, with regional disparities playing an important part (Adler & Ansell, 2020; Ballard-Rosa et al., 2021a, 2021b; Broz et al., 2021; Carreras et al., 2019).

Such economic distress may be a direct consequence of globalization, specifically of the profound redistributive effects of import competition (Autor et al., 2013, 2016). Recent research shows that, across Western Europe, parties with nationalist platforms have gained larger vote shares in regions ‘shocked’ by surging Chinese imports (Colantone & Stanig, 2018a; Dippel et al., 2015; Malgouyres, 2017). This line of research suggests that the nationalist backlash against globalization at the ballot boxes may partly be a direct consequence of the economic transformations caused by globalization itself. Yet, open questions remain as to the precise individual-level mechanisms underlying increasing vote shares for the radical right in regions affected by import competition.

In the present study, we transfer the idea that the nationalist backlash is caused by regional exposure to import competition to the level of individual political attitudes. Our key argument is that individuals living in regions suffering from import competition form more nationalist and isolationist attitudes as part of a broad counter-reaction to globalization. We thus expect them to become more emotionally attached to the nation and less supportive of transfers of political power from the national to the international level. We test this nationalist backlash hypothesis using panel data, which allow us to identify the effects of interests from intra-individual change in nationalist attitudes, thereby also contributing an important methodological innovation to research in this area.

Specifically, this study uses data from the British Household Panel Study (BHPS) to look at the evolvement of nationalist attitudes between 1999 and 2008. This period covers China’s WTO (World Trade Organization) entry in 2001 and the subsequent surge in imports from China—i.e., the ‘China shock’ (Autor et al., 2016). Our focus on imports from China is motivated by the consideration that rapidly increasing imports from China caused significant structural economic change in many developed countries (Autor et al., 2013, 2016). We leverage this ‘China shock’ as a clearly identifiable case to learn about the general phenomenon of low-wage import competition. Like previous studies, we capture changes in region-specific import competition by combining data on increases in Chinese imports across sectors with employment shares of these sectors at the NUTS (Nomenclature of Territorial Units for Statistics) 3 regional level.

Replicating results from previous contributions (Colantone & Stanig, 2018b) within our innovative panel data analysis, we find that individuals from British regions more exposed to Chinese import competition turn more critical of European Union (EU) membership. On top of that, we document a general increase in nationalist attitudes, with individuals becoming less tolerant of criticizing the nation and more opposed towards cooperation with other countries. We thus reveal a direct individual-level response to regional trade shocks in the form of rising nationalist attitudes that helps to explain their electoral consequences in terms of growing vote shares for parties of the radical right.

While our results corroborate findings of the existing literature on the electoral consequences of (Chinese) import competition, our study is innovative in at least three important respects:

First, while previous research (Autor et al., 2020; Colantone & Stanig, 2018a, 2018b; Dippel et al., 2015; Malgouyres, 2017) provides compelling evidence that local import shocks affect voting behavior, and that nationalist parties of the radical right—who oppose globalization most fiercely, particularly on the cultural dimension (Kriesi et al., 2008)—tend to profit most from local trade shocks, our study produces additional insights on the individual-level mechanisms that connect trade shocks and aggregate-level outcomes. Colantone and Stanig (2018b, p. 7) propose three mechanisms that may underlie their results on Chinese import competition and support for Brexit: Increasing opposition to incumbent political elites, increasing opposition to immigration, and a ‘nationalist and isolationist syndrome’. While they report empirical evidence supporting the second mechanism, the results of our study imply—without ruling out additional mechanisms—that vote shares of nationalist parties increase because nationalist attitudes increase among voters, i.e., the nationalist backlash observed at the polls is, at least to some extent, a result of a nationalist backlash in individual attitudes.

This is also what sets our study apart from parallel research that studies the impact of local trade shocks on political attitudes (Ballard-Rosa et al., 2021a, 2021b; Bisbee et al., 2020; Colantone & Stanig, 2018c; Ferrara, 2021; Hays et al., 2019). While these studies concentrate on attitudes towards immigration and minorities as well as authoritarian values, we study nationalist attitudes in the sense of an uncritical attachment to the nation and support for isolationism. In line with Colantone and Stanig (2018b, p. 7), we suggest that this may be an important part of the public’s response to import shocks. In fact, this conjecture accords with a key tenet of one of the hallmark paradigms in international political economy, the theory of embedded liberalism (Ruggie, 1982): The economic vagaries resulting from open markets may endanger not only public support for international economic openness (Hays et al., 2005), but for a multilateral world order more broadly.

The third innovation of our study is the use of individual-level panel data. Our empirical analysis crucially benefits from the fact that the attitudinal items of interest run repeatedly in the BHPS. For a clean identification, we leverage intra-individual changes in attitudes over time. This helps in overcoming the problem—inherent to cross-sectional designs—that import shocks might be correlated with initial differences across regions, be it in voting patterns or political attitudes (see Goldsmith-Pinkham et al., 2020). In fact, the variation in industry specialization across regions—from which the local import shock is computed—in itself may contribute to differences in political attitudes and behavior across regions, as people’s workplaces shape their political preferences (Kitschelt & Rehm, 2014). While previous and parallel research relying on (repeated) cross-sections draws on a range of techniques to carefully deal with that confoundedness concern, it remains a crucial advantage to observe intra-individual changes in political attitudes.

The remainder of this paper is structured as follows. In the next section, we elaborate upon the nationalist backlash hypothesis. The third section introduces our research design. The fourth section presents our empirical results. The final fifth section summarizes and discusses our key findings.

2. The nationalist backlash hypothesis

We begin our theoretical argument by setting it in relation to parallel papers that also study the impact of local trade shocks on political attitudes, but do so with a focus on different attitudes (Ballard-Rosa et al., 2021a, 2021b; Bisbee et al., 2020; Colantone & Stanig, 2018c; Ferrara, 2021; Hays et al., 2019). The common theoretical denominator across these studies is the notion that threats to and frustrations with one’s social status push demands for norm as well as cultural conformity. This goes along with increased authoritarian tendencies (Ballard-Rosa et al., 2021a, 2021b) and hostility towards out-groups, in particular immigrants (Bisbee et al., 2020; Colantone & Stanig, 2018c; Ferrara, 2021; Hays et al., 2019).

The expectation of a nationalist backlash is related to these ideas, as identification with the nation can be psychologically attractive for individuals who face threats to their social status. According to Shayo’s (2009) model of national identity, for example, the attractiveness of identifying with the nation increases when the psychological reward from alternative forms of group identification—such as class—erodes due to a decrease in social status of these groups. For individuals who face economic hardships and status threats, obedient attachment to the nation can thus be attractive. This tendency may be stronger to the extent that international competition is present, rendering ‘one’s membership in the nation a more salient attribute’ (Shayo 2009: 155).

Yet, we believe there is an even more straightforward mechanism—one that is compatible with the main thrust of such arguments—connecting exposure to import competition and a rise in nationalist and isolationist attitudes: The idea of a broad nationalist counter-reaction to globalization. The simple intuition behind this nationalist backlash hypothesis is that those negatively affected by one aspect of globalization, such as international trade, turn critical towards other facets of globalization, such as the transfer of political power from the national to the international level. It is an established proposition that individual material self-interest pushes those who are negatively affected by international trade to oppose it (Mayda and Rodrik 2005; Scheve and Slaughter 2001; Jäkel and Smolka 2017; Harms and Schwab 2020). Building on such findings, we reason the backlash caused by material consequences of trade not to be limited to attitudes towards trade, but to extend to a broader nationalist backlash against globalization in its various manifestations.

In making this argument, we assume that individuals do not neatly distinguish between the different facets of globalization, but that these objects are connected within their belief systems. This assumption finds justification in a sizable body of literature claiming that there is a new political divide, or even cleavage, around issues related to globalization or denationalization (de Vries 2018; Hooghe and Marks 2018; Kriesi et al. 2008). Studies within this literature reveal that individuals’ attitudes on different facets of globalization, such as immigration, European integration and free trade, are closely associated with each other (de Vries 2018; Hillen and Steiner 2020; Kriesi et al. 2008). Moreover, citizens, when repeatedly surveyed, report stable attitudes towards the concept of ‘globalization’ that are closely aligned with orientations towards such specific globalization-related issues, and independently affect voting decisions (Mader et al. 2020). Such evidence suggests that globalization may be an emotionally charged symbol for broad changes in the economic, social and political domain that individuals view in general either positively or negatively. As a result of these mental connections, when negatively affected by international trade, individuals are likely to think about globalization, in general, in more negative terms, and support for isolationism and nationalist sentiments is likely to increase.

In a similar vein, Margalit (2012, p. 487) proposes that ‘people view the material effects of trade as only one component of a broader “package” of openness that includes processes such as […] the increasing exposure to foreign influences [or] a shift towards a less traditionalist society’. Margalit invokes this mental connection to argue that nationalist and ethnocentric sentiments affect attitudes towards the prima facie economic issue of international trade. Our argument is that this logic works in the other direction as well in that the material consequences of trade may affect nationalist sentiments and views on the transfer of political power from the national to the international level. Thus, ‘the adverse material effects of openness to international trade [may] underlie […] the emergence of chauvinist-nationalist views’ (Margalit 2012: 485). Likewise, for Rodrik (2021) trade is inherently connected with national identity issues, because it presents an ‘obvious outsider target’.

As we focus on a region’s exposure to low-cost import competition, our argument rests on the assumption that the local context is important for individual attitude formation. Several channels might contribute to such a relevance of the local context (also see Broz et al. 2021). First, there might be direct as well as indirect effects of an increasing exposure to imports on individuals’ economic well-being, which then might have repercussions on political attitudes. Second, in addition to reacting to the (direct or indirect) effects on individual material well-being, individuals might be sociotropically motivated and care about their region of residence. Third, the local context might matter because communication among individuals could lead to a contagion of political attitudes.

In our empirical analysis, we thus expect individuals residing in regions subjected to growing import competition from China to turn more nationalist. Like previous work (Colantone and Stanig 2018b), we study support for EU membership as one important manifestation of this potential nationalist backlash. Going beyond EU support, we also test for a general increase in nationalist attitudes. This is crucial for an encompassing test of the nationalist backlash hypothesis and an important addition to the state of research.¹ By studying both EU support and general nationalist attitudes, we can assess whether growing EU skepticism is part of a broader nationalist backlash in response to import exposure, in line with our theoretical argument.

We thus test two hypotheses:

H1: Attitudes of individuals living in regions more exposed to growing Chinese imports become critical of EU membership.

H2: Attitudes of individuals living in regions more exposed to growing Chinese imports become more nationalist.

3. Data and methods

3.1. Dependent variables

We test these two hypotheses with data from the BHPS (University of Essex,  2010)—a nationally representative household panel running from 1991 to 2008. We measure nationalist attitudes from a battery of questions on ‘national identity’ administered in 1999, 2002, 2005 and 2008. Questions on support for EU membership were included in 1999, 2002 and 2006. These data thereby enable us to study changes in nationalist attitudes in the 2000s—i.e., during a period in which imports from China into the UK surged, following China’s entry into the WTO in 2001. To use this panel data efficiently, we study intra-individual changes in these attitudes in between two consecutive survey waves. Thus, the first observations pertain to changes between 1999 and 2002, and so forth.

Table 1 lists the items we use to measure our two main dependent variables. For attitudes towards the EU, we combine three questions that each ask about orientations towards British membership in the EU. Our (main) measure of a general nationalist attitude synthesizes three items on national pride, uncritical attachment to one’s nation and support for international cooperation.²

Table 1. Operationalization of nationalist attitudes in the BHPS.

Construct	Available years	Item wording	Scale	Loading on factor
Support for EU membership	1999, 2002, 2006	‘Generally speaking, do you think that Britain's membership of the European Union is a good thing, a bad thing or is it neither good nor bad?’	0–2	0.94
		‘Taking everything into consideration, would you say that Britain has on balance benefited or not from being a member of the European Union?’	0–1	0.91
		‘Do you think Britain's long-term policy should be … - to leave the European Union - to stay in the EU and try to reduce the EU's powers - to leave things as they are - to stay in the EU and try and increase the EU's powers or - to work for the formation of a single European government?’	0–4	0.82
Nationalist attitude	1999, 2002, 2005, 2008	‘I would rather be a citizen of Britain than of any other country in the world’	0–4	0.78
		‘People in Britain are too ready to criticize their country’	0–4	0.66
		‘Britain should co-operate with other countries, even if it means giving up some independence’	0–4	−0.51

Note: Factor loadings are from a polychoric principal component factor analysis.

We ran factor analyses—which returned acceptable factor loadings (see Table 1)—to combine these items into single scales using the predicted factor scores. Using observations from the overlapping years, the two resulting measures correlate with −0.37, indicating that EU support and nationalist attitudes are two distinct yet associated constructs. For our main dependent variables, we computed the first difference of each two consecutive observations. Change in EU support (nationalist attitude) has a mean of −0.09 (−0.01) and a standard deviation of 0.64 (0.87).

3.2. Model specification

Our statistical models study changes in individuals’ attitudes over time. We thereby aim to safeguard against the possibility that import shocks are correlated with initial differences in attitudes across regions. Within cross-sectional models it is hard to rule out that regions that later experienced larger import shocks were more nationalist to begin with, perhaps as a result of their larger share of manufacturing workers. Our models regress the first difference in the attitudes of interest in t on the lagged level value of this attitude in t-x and other covariates, with x indicating the number of years separating two consecutive observations. Our main specification is given by the following expression: (1) $\mathrm{\Delta }{Y}_{ir,t}=\alpha +\rho Y_{ir,t-x}+\beta C{S}_{r,t}+\sum _{k=1}^K{\gamma }_k{x}_{ir,t}^k+\sum _{l=1}^L{\delta }_l{x}_{r,t}^l+{\xi }_{r,t}+{\epsilon }_r+{\epsilon }_{r,t}+{\nu }_{ir,t}$(1) In (1), i indexes individuals, t years and r regions. We regress the first difference in an attitude of individual i residing in NUTS 3 region r at time t (Δ$Y_{ir,t}$) on its lagged value $Y_{ir,t-x}$, the corresponding Chinese import shock for the individual’s region of residence at time t ($C{S}_{r,t}$), a vector of individual-level controls ($x_{ir,t}^k$), a vector of regional-level controls ($x_{ir,t}^l$), and a set of NUTS 1 region-year fixed effects (${\xi }_{r,t}$).³

In addition to error terms at the individual-year level (${\nu }_{ir,t}$), equation (1)(1) $\mathrm{\Delta }{Y}_{ir,t}=\alpha +\rho Y_{ir,t-x}+\beta C{S}_{r,t}+\sum _{k=1}^K{\gamma }_k{x}_{ir,t}^k+\sum _{l=1}^L{\delta }_l{x}_{r,t}^l+{\xi }_{r,t}+{\epsilon }_r+{\epsilon }_{r,t}+{\nu }_{ir,t}$(1) includes error terms at the NUTS 3 regional level (${\epsilon }_r$) and the level of NUTS 3 region-year combinations (${\epsilon }_{r,t}$), treating both as random coefficients. This is crucial, as our data are characterized by a hierarchical three level structure (see Gelman and Hill 2006): Observations of individuals in year t (level 1) are nested within NUTS 3 region-year combinations (level 2), which are nested within NUTS 3 regions (level 3).⁴ Multilevel modelling is the proper way to address the inferential challenge that we are interested in how the Chinese import shock affects attitudes within a region, yet only observe a sample of individuals within each NUTS 3 region.⁵

3.3. Measuring local exposure to increasing Chinese imports

Our measurement of the Chinese import shocks at the level of NUTS 3 regions is based on the approach developed by Autor et al. (2013) and combines two pieces of information: The initial employment structure of a region and the increase in imports from China at the industry level. The idea is to infer how strongly regions are affected by increasing imports within a particular industry from the share of workers of a region that were initially employed in this industry. Simply, if many of the jobs within a region were in an industry that subsequently experienced a large increase in imports from China, the import shock will be high.

Formally, our main measure is defined as follows: (2) $C{S}_{r,t}=\sum _{j=1}^J{\omega }_{jr,t-x}\left({\displaystyle \frac{I{M}_{j,t}-I{M}_{j,t-x}}{I{M}_{j,t-x}}}\right)\ast 100$(2) with r indicating regions, j industries, and t standing for a given year. $I{M}_{j,t}$ is the real value (i.e., deflated by the Consumer Price Index, with 1995 used as base year) of UK imports in Pound Sterling from China in industry j. ${\omega }_{jr,t-x}$ are industry employment shares, i.e., the ratio of the number of workers in region r and industry j at time t-x to the total number of workers in region r in that period $\left({\displaystyle \frac{L_{jr,t-x}}{L_{r,t-x}}}\right)$. Equation (2)(2) $C{S}_{r,t}=\sum _{j=1}^J{\omega }_{jr,t-x}\left({\displaystyle \frac{I{M}_{j,t}-I{M}_{j,t-x}}{I{M}_{j,t-x}}}\right)\ast 100$(2) computes the growth rates (i.e., the percentage change) of the real value of imports from China in industry j between year t-x and year t, weights these growth rates by the share of employment for an industry in a region in the base year t-x, and then takes the sum over all industries within region j. The regional import shock thus defined is a weighted average of growth in imports per sector, with the weights being sectoral employment shares.⁶

We computed the China shock measure such that it corresponds to the time structure of our survey data, with t-x being the year in which the lagged attitude was measured. For instance, for predicting change in nationalist attitudes between 2008 and 2005, t is 2008 and t-x is 2005. We believe that, for predicting the change in an attitude, the change in exposure to Chinese imports over the same period (and based on the employment shares at the beginning of this period) is the most relevant quantity. In the appendix, we consider measures with alternative base years.

Our regional units are the 128 NUTS 3 regions in Great Britain according to the 2006 NUTS revision. To assign individual-years to those regions, we rely on a (conditional access) BHPS dataset on the local authority districts (LADs) that households are situated in (University of Essex,  2014). We computed the ‘China shock’ according to equation (2)(2) $C{S}_{r,t}=\sum _{j=1}^J{\omega }_{jr,t-x}\left({\displaystyle \frac{I{M}_{j,t}-I{M}_{j,t-x}}{I{M}_{j,t-x}}}\right)\ast 100$(2) , distinguishing between 21 harmonized industries in the primary and secondary sectors based on regional employment share data from NOMIS and industry-level data on imports from China from the OECD STAN database.⁷

Descriptive results on the China shock across regions in appendix section B.2 indicate face validity, but also severe skewness. To ensure that our results are not driven, or distorted, by a few heavy outliers, we logarithmized the original values, calculating: (3) $C{S}_{r,t}^{log}=ln(C{S}_{r,t}+1)$(3) For the nationalist attitude regressions, the mean value of $C{S}_{r,t}^{log}$ is 2.30—corresponding to $C{S}_{r,t}=8.97$, that is, a weighted average growth rate of Chinese imports of about 9 percent—and its standard deviation is 0.51. For the EU support regressions, the mean is 2.57 with a standard deviation of 0.49.

3.4. Control variables

As individual level control variables, we include gender, age (and age squared), education and immigration background (see appendix A.1 for further information on all control variables). At the regional level, we include fixed effects at the level of NUTS 1 region-year combinations, as described in equation (1)(1) $\mathrm{\Delta }{Y}_{ir,t}=\alpha +\rho Y_{ir,t-x}+\beta C{S}_{r,t}+\sum _{k=1}^K{\gamma }_k{x}_{ir,t}^k+\sum _{l=1}^L{\delta }_l{x}_{r,t}^l+{\xi }_{r,t}+{\epsilon }_r+{\epsilon }_{r,t}+{\nu }_{ir,t}$(1) . We thereby identify the effects of interest from variation across NUTS 3 region-years within NUTS 1 region-years. There are eleven NUTS 1 regions observed in our data, which include at least five (London) and up to 19 NUTS 3 regions (Scotland). In addition, we control for the employment share of manufacturing in 1998 of NUTS 3 regions. As increasing imports mainly affect manufacturing, this is a tough control, which is strongly correlated with the China shock (see appendix section C.3). Nonetheless, we opt to disentangle the specific impact of the China shock from developments common to traditional manufacturing regions, that possibly could have become more nationalist for other reasons than growing import exposure. We keep the value fixed at its 1998 level to avoid post-treatment bias, as changes in the manufacturing share over our observation period might be driven by rising imports from China. Furthermore, in some models we incorporate the share of the population born outside of the UK, including both levels and changes (in percentage points) between t-x and t, and measured at the level of Local Area Districts (LADs). This introduces another layer to the data structure. Accordingly, we add random intercepts at the level of LADs-years to the respective models such that these contain four nested levels (individual-years, LADs-years, NUTS 3 region-years, NUTS 3 regions).

4. Results

We present our main regression results in two tables. Table 2 studies change in EU support and Table 3 change in nationalist attitudes. In each table, we report results from four different specifications. The first three models introduce control variables at the regional level stepwise. Model (1) includes only NUTS 1 region-year fixed effects. Model (2) adds the 1998 manufacturing employment share. Model (3) additionally takes the level of and the change in the share of the population born outside of the UK into account. Model (3) is our preferred specification, and we use it as a baseline for (most of) our robustness checks reported below. Model (4) departs from our preferred specification, as described by equation (2)(2) $C{S}_{r,t}=\sum _{j=1}^J{\omega }_{jr,t-x}\left({\displaystyle \frac{I{M}_{j,t}-I{M}_{j,t-x}}{I{M}_{j,t-x}}}\right)\ast 100$(2) , in favor of a still more conservative version that includes fixed effects at the level of NUTS 2 (rather than NUTS 1) region-years.

Table 2. Regressing change in support for EU membership on Chinese import shock.

	(1)	(2)	(3)	(4)
Chinese import shock (log)	−0.043*	−0.064*	−0.065*	−0.036
	(0.017)	(0.028)	(0.028)	(0.035)
EU supportt-3/4	−0.37***	−0.37***	−0.37***	−0.37***
	(0.011)	(0.011)	(0.011)	(0.011)
EU supportt-3/4 X year=2006	−0.040**	−0.040**	−0.040**	−0.042**
	(0.015)	(0.015)	(0.015)	(0.015)
Manufacturing share1998		0.18	0.17	0.047
		(0.18)	(0.18)	(0.21)
Foreign born population			0.019	0.046
			(0.17)	(0.20)
Change in foreign born pop.			−0.53	−0.40
			(0.43)	(0.44)
Sociodemographic controls	☑	☑	☑	☑
Fixed effects
NUTS 1-Year	☑	☑	☑
NUTS 2-Year				☑
Observations
NUTS 3	122	122	122	122
NUTS 3-year	240	240	240	240
LAD-year			663	663
Individual-year	9556	9556	9556	9556
BIC	16261.8	16270.0	16292.1	16693.4
Standard deviation in estimation sample
Δ support for EU membership	0.63	0.63	0.63	0.63
Chinese import shock	0.49	0.49	0.49	0.49

Note: Results from linear multilevel models. Standard errors in parentheses. Full regression table is in appendix section C.2. Significance levels: + p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001.

Table 3. Regressing (change in) nationalist attitude on Chinese import shock.

	(1)	(2)	(3)	(4)
Chinese import shock (log)	0.039**	0.049*	0.048*	0.061*
	(0.014)	(0.019)	(0.020)	(0.027)
Nationalist attitudet-3	−0.52***	−0.52***	−0.52***	−0.52***
	(0.0056)	(0.0056)	(0.0056)	(0.0056)
Manufacturing share1998		−0.099	−0.077	−0.16
		(0.13)	(0.14)	(0.17)
Foreign born population			0.21	0.25
			(0.14)	(0.16)
Change in foreign born pop.			−0.20	−0.25
			(0.36)	(0.36)
Sociodemographic controls	☑	☑	☑	☑
Fixed effects
NUTS 1-Year	☑	☑	☑
NUTS 2-Year				☑
Observations
NUTS 3	123	123	123	123
NUTS 3-year	366	366	366	366
LAD-year			1039	1039
Individual-year	24159	24159	24159	24159
BIC	54304.1	54313.6	54332.2	54987.5
Standard deviation in estimation sample
Δ nationalist attitude	0.86	0.86	0.86	0.86
Chinese import shock	0.51	0.51	0.51	0.51

Note: Results from linear multilevel models. Standard errors in parentheses. Full regression table is in appendix section C.2. Significance levels: ⁺p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001.

Throughout Table 2, we obtain a negative coefficient of the China shock on change in support for EU membership. Thus, EU support decreases with increasing local exposure to growing Chinese imports. This effect is statistically significant in all three models with NUTS 1 region-year fixed effects, including our preferred specification in model (3) with the full set of regional-level controls. With the NUTS 2 region-year fixed effects in model (4), the coefficient remains largely consistent with the previous estimates, but does not reach statistical significance. We note that model (4) is quite demanding as it identifies the effect of interest only from variation across NUTS 3 region-years within NUTS 2 region-years—and there are few NUTS 3 regions per NUTS 2 region.⁸ Overall, however, these baseline results are in line with the expectation of a negative effect of the Chinese import shock on EU support. In contrast, none of our regional-level controls turn out statistically significant.

The coefficient in our preferred specification in Table 2, model (3), indicates that a one-unit increase in the logarithmized import shock leads to a decrease in EU support by 0.065. This corresponds to roughly a tenth of a standard deviation in the observed first difference in EU support (=0.63).

To give a better substantive sense of this effect, consider that a one-unit increase in the logarithmized import shock from 2.3 to 3.3 means, for example, moving from a weighted growth rate of sectoral Chinese imports of about 9 percent to 26 percent, with the former value corresponding to the 23th percentile and the latter value to the 89th percentile of the import shock. Over the period from 2002 to 2006, we observe related values for Outer London - East and North East (import shock of 9.38) and Solihull in the West Midlands (import shock of 27.0), for example. Thus, the model predicts EU support to decrease by roughly 0.065 scale points between 2002 and 2006 among respondents from Solihull relative to individuals living in East Outer London (net of other variables in the model).⁹

While this is not a huge effect, it is substantially meaningful, given that this is an estimate of an (average) effect for all individuals within NUTS 3 regions. When interpreting these effect sizes, we also need to bear in mind that we are studying consecutive changes over a small number of years. From a long-time perspective, the effects might cumulate over time. Even over the window we observe in the models in Table 2, the China shock hits two times (from 1999 to 2002 and from 2002 to 2006), with the effects potentially adding up.

These results so far are well in line with Colantone and Stanig’s (2018b) findings on local Chinese import shocks and support for Brexit. Yet note that we replicate this finding within a very different research design. Most notably, our results indicate that changes in individuals’ attitudes towards EU membership between 1999 and 2006 are associated with exposure to the sectoral growth of Chinese imports during this period. Colantone and Stanig’s (2018b) estimations instead associate levels of leave voting in the June 2016 Brexit referendum with local variation in exposure to increasing Chinese imports at the NUTS 3 level between 1990 and 2007.

Our next set of findings are novel in more obvious ways, in that we study whether there is also a general nationalist shift in political attitudes. In Table 3, we present results from an analogous set of regression models for the measure of change in nationalist attitudes. Throughout the four specifications we observe a statistically significant positive effect of local exposure to growing Chinese imports. The coefficients are of similar magnitude across the four specifications yet tend to get larger in more saturated models while the standard errors increase slightly. Overall, there is strong evidence that exposure to growing Chinese imports at the NUTS 3 level goes along with an increase in nationalist attitudes. In contrast, none of the regional-level control variables are statistically significant.

In terms of the substantive magnitude of the estimated effects of the import shock, the results are in a similar order of magnitude as those for EU support. The coefficient in our preferred specification in Table 3, model (3), indicates that a one-unit increase in the logarithmized import shock leads to an increase in nationalist attitudes of 0.048. Again, it is instructive to relate this value to observed values in the untransformed measure of the import shock: A one-unit increase in the logarithmized import shock from 2.04 to 3.04 means, for example, moving from a weighted growth rate of sectoral Chinese imports of about 6.7 percent to 20 percent. This corresponds to moving from the 30th percentile to the 95th percentile of the import shock observed in model (3). Over the period from 2002 to 2005, we observe an import shock of 6.88 for Outer London - East and North East and an import shock of 20.2 for Coventry in the West Midlands, for example. Thus, the model predicts nationalist attitudes among respondents from Coventry to shift in a more nationalist direction from 2002 to 2005 by roughly 0.048 scale points as compared to individuals living in East Outer London (net of other variables in the model).

Again, this is not an especially large effect, yet it does indicate a substantially meaningful one. Especially once we take into account that this is an estimate of an (average) effect for all individuals within NUTS 3 regions and one that applies to changes in nationalist attitudes over a three-year window only. Given that such effects might cumulate over a longer time span, including the three intervals covered by our models, sizable differences in levels of nationalist attitudes across regions might be the consequence.

Overall, we obtain support for the nationalist backlash hypothesis. Replicating previous results on the Brexit referendum in a different set-up, our results indicate that EU support has decreased over time among individuals living in regions exposed to growing Chinese imports due to their industry specialization. We have argued that this shift is part of a broader nationalist backlash. The novel evidence on general nationalist attitudes—composed of nationalist sentiment and opposition to international cooperation—is in line with this conjecture: We find that attitudes shift in a nationalist direction among individuals residing in regions affected by growing Chinese imports.¹⁰

We extended our baseline analyses in several directions (see the appendix). Section C.4 shows that our results are robust to different choices as to how to measure nationalist attitudes—either through single items or alternative scales. In section C.6, we report results from an extensive set of robustness checks. When using a constant initial base year against which to compute the growth in imports, our results remain also largely similar, though it seemingly becomes harder to disentangle the effect of import exposure from the effect of the initial employment share in manufacturing. As an alternative to our preferred baseline multilevel model, we estimated individual-level fixed effect models. Although identification is based only on over time-variation in regional exposure in these models and most of the variation in the China shock is between NUTS 3 regions rather than within NUTS 3 regions over time, we continue to find statistically significant positive effects on nationalist attitudes. For EU support—where we only have three observations over time—the coefficient fails to reach statistical significance, though the negative coefficient we obtain is consistent with the baseline findings. Other checks include, for example: The exclusion of movers, and of workers in the primary and secondary sectors; and instrumental variable estimations, using Chinese exports to other countries as an instrument. Overall, these additional estimations attest to the robustness of our main findings, especially for the novel finding that regional exposure to import competition is associated with an increase in general nationalist attitudes.

In section C.7, we report results from models with indicators of changes in local economic activity. The findings suggest that the nationalist backlash is a specific reaction to import competition, not merely a general consequence of local economic decline (cf. Colantone and Stanig 2018a: 211). In section C.9, we interact import exposure with individual characteristics, obtaining results largely in line with a sociotropic reaction. In section C.10, we add a measure of rising exposure to growing Chinese imports at the level of sectoral employment. We find similar, albeit weaker, results for this occupational import shock—in line with our general argument that exposure to import competition causes a nationalist backlash in political attitudes. The regional effect remains intact in these regressions, supporting our interpretation of a genuine local effect that exists independently from—and possibly in addition to—affectedness by virtue of being oneself employed in an industry facing growing import competition.

In section D, we report results from a similar set of regressions with the dependent variable being indicators of attitudes on economic policy following work on the ‘compensation hypothesis’ and its individual-level mechanics. In stark contrast to the robust effects we saw on EU support and nationalist attitudes, we do not obtain any supportive evidence that economic policy orientations shift leftwards in response to exposure to growing Chinese imports.

5. Conclusion

This paper has addressed the question whether intensifying exposure to low-wage import competition at the regional level induces individuals to adopt an increasingly nationalist attitude. Answering this question is important for understanding the sources behind the anti-globalization backlash in Western democracies. While previous studies have provided evidence that exposure to import competition contributes to the success of nationalist parties, our study of the consequences of import shocks has studied political attitudes directly and serves to better understand why we observe these effects on voting behavior. To present clean evidence on this matter, we combined data on regional exposure to the surge in imports from China with panel data from the BHPS, focusing on changes over time for identification.

The results, including a range of robustness checks, are broadly supportive of our nationalist backlash hypothesis. To begin with, our findings corroborate the findings of Colantone and Stanig (2018b) on the Brexit referendum in that we find regional exposure to Chinese imports to be associated with growing opposition to EU membership within a different, longitudinal research design. Importantly, we have provided novel evidence that this effect is not limited to attitudes towards the EU but extends to a general nationalist backlash in political attitudes. Individuals living in regions affected by growing Chinese imports by way of their sectoral specialization tend to exhibit an increase in nationalist sentiment and a growing skepticism towards international cooperation. These results are in line with the nationalist backlash argument we proposed: Individuals negatively affected by globalization in terms of low-cost import competition view globalization in general more critical, become more attached to the nation and supportive of isolationist stances.

In stark contrast, we obtained no evidence for a leftward shift in economic policy positions. This pattern of results helps us better understand why it is mostly not left parties who profit from attracting globalization losers, but parties of the nationalist right (Colantone and Stanig 2018a). Our findings are thus in line with the central tenet of the theory of embedded liberalism that compensating globalization’s losers might be necessary to sustain public support for an open world order based on multinational cooperation. At the same time, we do not observe that losers from international trade demand such compensation. They seem to rather turn against globalization and multilateralism itself.

One obvious limitation is that our results are from a single country. There are reasons to expect relatively pronounced effects of import competition in the British case, given limited welfare state compensation and large economic disparities across regions. Future research may, inter alia, investigate—perhaps building on our study design—how the nationalist backlash in political attitudes is sensitive to such context conditions.

Statistical replication materials and data

Supporting data and materials for this article can be accessed at https://doi.org/10.7910/DVN/OYD4AQ.

Acknowledgements

We are indebted to Mark Ruszel for excellent research assistance. Previous versions of this manuscript were presented in seminars at JGU Mainz and the University of Bayreuth, the EPSA 2018 conference in Vienna, the DVPW 2018 conference in Frankfurt, the 2019 meeting of the DVPW section on Political Economy in Konstanz, the ECPR 2019 conference in Wrocław, the EPCS 2019 conference in Jerusalem, the EEA 2019 conference in Manchester, the ETSG 2019 conference in Berne, and the Verein für Socialpolitik 2019 conference in Leipzig. We are grateful for the many helpful comments and suggestions we received along the way. Special thanks go to JEPP’s anonymous reviewers for their thoughtful and constructive remarks.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

Financial support for this research was provided by the Research Unit ‘Interdisciplinary Public Policy (IPP)’ at Johannes Gutenberg University Mainz.

Notes on contributors

Nils D. Steiner

Nils D. Steiner is a post-doctoral researcher at the Department of Political Science, Johannes Gutenberg University Mainz, Germany.

Philipp Harms

Philipp Harms is a Professor of International Economics at the Gutenberg School of Management and Economics, Johannes Gutenberg University Mainz, Germany.

Notes

1 Note that data limitations prevent us from studying how attitudinal changes brought about by import exposure eventually affect voting behavior. While we aim to speak to a broader debate on the electoral consequence of import exposure, our specific contribution to it is focused on the link between import exposure and nationalist attitudes.

2 These are admittedly somewhat heterogenous items, tapping into isolationism and different facets of nationalism—‘national chauvinism’, the belief in the inherent superiority of one’s nation (Herrmann 2017: S70), and ‘blind patriotism’, the ‘attachment to country characterized by unquestioning positive evaluation, staunch allegiance, and intolerance of criticism’ (Schatz et al. 1999: 151). In robustness checks, we considered alternative scales, including additional items, and found our results to be robust (see below).

3 Note that this model is equivalent to a standard lagged dependent variable model which has $Y_{ir,t}$ rather than $\mathrm{\Delta }{Y}_{ir,t-x}$ as a dependent variable. We chose this version to emphasize that this is a model of change. Controlling for the lagged level value in t-x is essential to capture regression-to-the-mean effects, which naturally arise from response scales with end points (see appendix section B.3). For example, if someone already scores maximally high on nationalist attitudes, nationalist attitudes cannot increase any further. Because EU support was included in irregular intervals, we additionally included interactions between the lagged level and the year dummies in the respective regressions.

4 Results are almost perfectly identical when we omit the third level.

5 In fact, we believe that, in our setting, ignoring the multilevel structure is a bigger concern than a potential endogeneity of Chinese imports to the UK. For this reason, we report results from an instrumental variable estimation in the robustness check only (see below) and report results from multilevel models as our main results.

6 Note that this measure slightly differs from Autor et al. (2013), which is based on the increase in imports in real Pound Sterling per worker in industry j. We utilize this alternative for two reasons. First, we suspect that growth rates capture the processes underlying the attitudinal response to import shocks in a better way: They take high values if industries facing little import competition in the past experience surging imports. In these situations, import competition is especially likely to be perceived as a growing threat and, hence, likely to trigger an attitudinal response. Second, the growth rate measure accounts for industries’ ‘initial labor market relevance’ (see appendix section A.3). Our main results are similar when we use a measure based on the increase in imports per worker (see appendix section C.3).

7 We provide more information on data compilation and descriptive statistics in the appendix. Due to lacking regional employment data, Northern Ireland could not be included.

8 We observe 36 NUTS 2 regions, which include between one and eight NUTS 3 regions. Of course, NUTS 2 regions which include only one NUTS 3 regions do not contribute to the estimate of the China shock in model (4) at all.

9 We chose this comparison to illustrate the effect of Chinese import competition for a specific NUTS 3 region pair. Since equation (3)(3) $C{S}_{r,t}^{log}=ln(C{S}_{r,t}+1)$(3) implies a non-linear mapping of $C{S}_{r,t}$ into $C{S}_{r,t}^{log}$ (and vice versa), the difference in $C{S}_{r,t}$ associated with a one-point difference in $C{S}_{r,t}^{log}$ is lower (higher) for region pairs at the lower (upper) end of the distribution.

10 In our argument, decreasing EU support and increasing nationalist attitudes are part of the same anti-globalization nationalist backlash, and as such are simultaneously determined. Alternatively, decreasing EU support might be thought of as a consequence of rising nationalism. In appendix section C.8, we add the change in nationalist attitudes to the EU support regressions. In line with such (partial) mediation, we find an increase in nationalist attitudes to predict a decrease in EU support. At a minimum, this shows that both attitudes move in tandem.